Purpose Few features characterizing the dosimetric properties of the patients are included in currently available dose‐volume histogram (DVH) prediction models, making it intractable to build a correlative relationship between the input and output parameters. Here, we use planning target volume (PTV)‐only treatment plans of the patients (i.e., the achievable dose distribution in the absence of organs‐at‐risk (OAR) constraints) to estimate the potentially achievable quality of treatment plans and establish a machine learning‐based DVH prediction framework with the use of the dosimetric metric as model input parameters. Methods A support vector regression (SVR) approach was used as the backbone of our machine learning model. A database containing volumetric modulated arc therapy (VMAT) plans of 63 prostate cancer patients were used. For each patient, the PTV‐only plan was generated first. A correlative relationship between the OAR DVH of the PTV‐only plan (model input) and the corresponding DVH of the clinical treatment plan (CTP) (model output) was then established with the 53 training cases. The prediction model was tested by the validation cohort of ten cases. Results For the training cohort, the checks of dosimetric endpoints (DEs) indicated that 52 of 53 plans (98%) were within the 10% error bound for bladder, and 45 of 53 plans (85%) were within the 10% error bound for rectum. In the validation tests, 92% and 96% of the DEs were within the 10% error bounds for bladder and rectum, respectively, and eight of ten validation plans (80%) were within the 10% error bound for both the bladder and rectum. The sum of absolute residuals (SAR) achieved a mean of 0.034 ± 0.028 and 0.046 ± 0.021 for the bladder and rectum, respectively. Conclusions A novel dosimetric features‐driven machine learning model with the use of PTV‐only plan has been established for DVH prediction. The framework is capable of efficiently generating best achievable DVHs for VMAT planning.

No abstract

An accurate prediction of achievable dose distribution on a patient specific basis would greatly improve IMRT/VMAT planning in both efficiency and quality. Recently machine learning techniques have been proposed for IMRT dose prediction based on patient’s contour information from planning CT. In these existing prediction models geometric/anatomic features were learned for building the dose prediction models and few features that characterize the dosimetric properties of the patients were utilized. In this study we propose a method to incorporate the dosimetric features in the construction of a more reliable dose prediction model based on the deep convolutional neural network (CNN). In addition to the contour information, the dose distribution from a PTV-only plan (i.e. the plan with the best PTV coverage by sacrificing the OARs sparing) is also employed as the model input to build a deep learning based dose prediction model. A database of 60 volumetric modulated arc therapy (VMAT) plans for the prostate cancer patients was used for training. The trained prediction model was then tested on a cohort of ten cases. Dose difference maps, DVHs, dosimetric endpoints and statistical analysis of the sum of absolute residuals (SARs) were used to evaluate the proposed method. Our results showed that the mean SARs for the PTV, bladder and rectum using our method were 0.007  ±  0.003, 0.035  ±  0.032 and 0.067  ±  0.037 respectively, lower than the SARs obtained with the contours-based method, indicating the potential of the proposed approach in accurately predicting dose distribution.

No abstract